Redemption in the Face of Stale Criminal
Records Used for Background Checks
Alfred Blumstein
Carnegie Mellon University
Kiminori Nakamura
University of Maryland
September 13, 2011
Partially supported by NIJ Grant No. 2007-IJ-CX-0041 and No. 2009-IJ-CX-0008.
Opinions or points of view expressed are those of the authors and do not necessarily
1
reflect the official position or policies of the U.S. Department of Justice.
Motivation for the Project
 Technology has made background checking easy –
and so very ubiquitous
 Most large companies (80-90%) now do background checks
 Statutes require background checks for many jobs and
occupational licenses
 Criminal records are also ubiquitous
 Nearly 14 million arrests a year
 92 million criminal records in state repositories
 As a result, many people can’t get a job or otherwise
– all because of a crime that happened long ago
 Need to explore when relief from that prior mark of
crime – “redemption” – should be granted
2
3
4
The Problem of “Redemption”
 It is well established that recidivism probability declines
with time clean after an arrest or conviction
 If a person with a criminal record remains crime-free
sufficiently long, his risk becomes less than some
appropriate comparison groups
 That record is then “stale”
 Now have some strong empirical estimates of when
redemption is appropriate - “redemption time”
5
Research Approach
 Data: Arrest history records (“rap sheets”) from NY
state criminal-history repository (DCJS)
 All individuals who were arrested for the first time in NY as
adults in 1980 (≈ 88,000)
 Follow-up time > 25 years
 Focus on a subset of arrestees who were convicted
 Age at first arrest: A1 = 19-20 vs. 25-30
 Crime type of first arrest: C1 = Violent vs. Property vs. Drug
 Measure of Recidivism Risk: Hazard
 Probability that a new arrest occurs at any particular time t
for those who stayed clean until t
 New arrest here could be for any crime type (C2 = any)
 Will later consider concern about specific subsequent crime types 6
Two Comparison Criteria
 Compare hazard of those with a prior to
1) Risk of arrest for the general population of the
same age: T*
2) Risk of arrest for those with no prior arrest: T**
7
T*: Comparison to General
Population of the Same Age
 Comparison based on the age-crime curve:

A(a) = rate of arrest at age a in the general population
 Estimate time to redemption, T*, when hazard
drops below the age-crime curve
8
Estimation of T*
T* = 3.8 years
Hazard at T* = .10
.20
.18
Probability of Arrest
.16
.14
.12
Age 19-20 Violent
.10
.08
Age-crime curve (General
population)
.06
.04
.02
.00
0
1
2
3
4 5 6 7 8 9
Years Since First Arrest
10 11 12
9
T**: Comparison to the
Never Arrested
 Comparison between the risk for those with a prior vs.
those without a prior arrest (the never arrested)
 Simple intersection method used for T* won’t work if the
never arrested is consistently below hazard
 Estimate time to redemption, T**, when hazard and the
never arrested are “close enough”
10
T**: Comparison to the
Never Arrested
.20
.18
Probability of Arrest
.16
.14
The never arrested
.12
.10
Age 19-20 Violent
.08
.06
?
.04
?
.02
?
.00
0
5
10
Years Since First Arrest
15
20
 Introduce a risk tolerance (“close enough”)
 How much extra risk an employer is willing to tolerate – perhaps
11
to get a better employee
Estimation of T**
.25
T** = 12.6 years
Hazard at T** = .03
.20
Probability of Arrest
Margin of error
.15
The never arrested
Age 19-20 Violent
Lower CI
.10
Upper CI
δRisk
= .02tolerance
.05
Risk tolerance
.00
0
5
10
Years since First Arrest
15
20
12
Choice of Benchmark: T* or T**?
 Factors to be considered:
 Applicant pool
 Many with priors (T*) vs.
 Primarily never-arrested (T**)
 Nature of the job and its risk sensitivity
 < 10% risk (T*) for minor theft from cash register or bar-room
fight
 < 3% risk (T**) for embezzlement risk or assault of vulnerable
customers
13
Concern for Arrests Outside NY
 Those who appear clean in NY might have been
arrested elsewhere
 Obtained from the FBI national criminal records for a
sample of 1980 NY arrestees with no re-arrest in NY
 About 23% of them were found to have arrests elsewhere
 Adjusted hazard for out-of-state arrests based on
 The proportion of the FBI sample with out-of-state arrests
 The distribution of time to the out-of-state arrest
14
Concerns about Robustness
 Estimates of redemption times are based on 1980
first-time arrestees in NY
 How reliable are our estimates for use at different
times or in different places?
 We test the robustness of estimates to:
 Different Sampling years (‘85, ‘90 from NY)
 Different States (Florida, Illinois in 1980)
15
Hazards across Three Sampling Years
A1 = 19-20
C1 = Violent
Probability of Rearrest
.30
.25
.20
1980
.15
1985
.10
1990
.05
.00
1
3
5
7
9
11
13
A1 = 19-20
C1 = Property
Probability of Rearrest
.30
.25
.20
1980
.15
1985
.10
1990
.05
.00
1
3
5
7
9
Years Since First Arrest
11
13
16
Hazards from NY, FL, and IL
.40
A1 =19-20
C1= Violent
Probability of Rearrest
.35
.30
.25
NY
.20
FL
.15
IL
.10
.05
.00
.40
0
5
10
15
A1 =19-20
C1= Property
Probability of Rearrest
.35
.30
.25
.20
NY
.15
FL
.10
IL
.05
.00
0
5
10
Years Since First Arrest
15
17
Robustness of Redemption Times
 We are reasonably comfortable with the robustness of our
hazard estimates
 Less so for the first 5-year interval, but that period is less relevant for
consideration of redemption
 We find them to be much more similar in later years
 Now we examine how these hazard differences affect the
redemption times
 Two general benchmark probabilities for redemption
 Probability of re-arrest for the 1980 cohort
Benchmarks
 T* ≈ .1
(probability of re-arrest)
 T** ≈ .03
C1
.1
.03
Violent
4-7
11-15
Drugs
4
10-14
Property
3-4
8-11
18
Concern over C2 – The “Next Crime”
 Employers differ in the crime types they care more
about
 Shop owners or banks care more about property crimes
 Those dealing with vulnerable populations care more
about violence
 EEOC requires employers to demonstrate “business
necessity” to justify the use of criminal records
 The future concern based on the prior record should be
job related
19
Approaches
 Develop crime-switch matrix based on arrests
 Probability of going from C1 to any particular C2
 Crime-type-specific hazards of 2nd arrest
 Conditional probability of a new arrest for particular
crime types (C2)
 C2-specific redemption times
 Use age-crime curves for particular C2
20
Crime-Switch Matrix for C1 to C2
Row Percents
C2
C1
Violent
Property
Drugs
Violent
Property
Drugs
Public Order
Others
21
9
8
12
10
14
26
11
11
14
9
6
24
7
6
Public
Order
8
6
6
22
6
Others
7
6
4
5
14
No
Diag/Avg of
2nd arrest Off-Diags
40
2.2
47
3.9
47
3.3
42
2.5
1.6
50
Avg = 2.7
21
C2 -Specific Hazard & T* Estimates
(A1 = 19-20, C2 = Violent)
Probability of Rearrest for Violent
.09
.08
.07
C1:
.06
.05
Violent
.04
Property
.03
Drugs
.02
.01
.00
0
5
10
Years Since First Arrest
C1
Violent
Drugs
Property
15
20
T*
14.0
5.2
4.6
22
Probability of Rearrest for Property
C2-Specific Hazard & T* Estimates
(A1 = 19-20, C2 = Property)
.12
.10
.08
C1:
.06
violent
property
.04
drugs
.02
.00
0
5
10
Years Since First Arrest
C1
Violent
Property
Drugs
15
20
T*
6.1
5.2
4.8
23
Redemption Policies
 Employers
 Inform employers of the low relevance of stale records
 Enact statutes to protect employers from “due-diligence liability”
claims – especially if they accept reasonable risk tolerance
 Redemption time should not interfere with reentry support –
Employment should be facilitated as soon as possible
 Especially with employment situations that are risk tolerant
(e.g., construction)
 Other information should be used to encourage employment
(e.g., positive work history, marriage)
 Repositories & commercial vendors
 State repositories could withhold stale records
 Could seal (or perhaps expunge) sufficiently stale records
24
Conclusions
 Recidivism risk declines with time clean
 Important consideration to many employers
 Redemption times (T* and T**) identify key time points
when the criminal record loses its value in predicting risk
 Strong empirical estimates of redemption times
 Based on a large set of official data
 Tested for robustness over time and across states
 Other researchers have produced similar estimates (Kurlychek et al.
2006, 2007; Bushway et al. 2011; Soothill and Francis 2009)
 Provides a basis for responsiveness to user criteria in assessing
redemption
 Redemption times can be generated based on the specifications (A1,
C1, δ, C2, etc.) set by the users
 Avoids inappropriately denying jobs to people with stale records
25
Future Work
 Consider more complex prior records
 Greater prior involvement in crime is associated with a higher risk of
recidivism
 Those with multiple prior records are more likely to be incarcerated for a
reasonable length of time
 Need better data on “time served”
 Special case of prison releasees
 Prejudice against ex-prisoners
 Could well have longer records
 Warrant redemption also, probably with longer redemption times
 Still need estimates of their redemption times based on sample of releasees
 Take account of treatments/programs while incarcerated and in community
 Applying the redemption methods to determine the timing of parole
discharge
 How long does it take for parolees’ recidivism risk to decline to a level that
no longer warrants the level of supervision required when newly released?
26
Thank you!
Questions & Suggestions?
27